Thrilled to introduce Foundation Model Self-Play, led by Aaron Dharna. FMSPs combine the intelligence & code generation of foundation models with the curriculum of self-play & principles of open-endedness to explore diverse strategies in multi-agent games. Thread x.com/jeffclune/st...

Nice summary, with coverage of the Darwin Gödel Machine. www.youtube.com/watch?v=C1ku...

A great summary of the latest in open-endedness research! The next big wave in AI? 🤔📈🚀 richardcsuwandi.github.io/blog/2025/op...

How it felt to know AGI is coming soon long before the world was paying attention. officechai.com/ai/building-...

Very nice summary of the Darwin Gödel Machine in Fortune, and an interesting tie-in to Sam's excellent recent blog post. fortune.com/2025/06/19/o...

I am excited to be a part of @Yoshua_Bengio's new non-profit focused on AI Safety and Existential Risk, joining the great team of Scientific Advisors. This is a critically important mission for humanity.

It's Friday Night, which means Trump's weekly science bloodbath. BUT WHY DOES HE DO THIS ON FRIDAY NIGHTS? Simple: He's a coward! Americans overwhelmingly support science! His attack on science is unpopular and he wants to bury it in the news cycle. Not on our watch. #StandUpForScience

Excited to introduce the Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents. We harness the power of open-ended algorithms to search for agentic systems that get better at coding, including improving their own code.

Is there a cancer at the heart of modern AI, lurking just beneath the surface of its dazzling performance? Our research suggests maybe, but also shows elegant solutions are possible (though how to get them at scale remains a mystery). Check out the eye-opening, riveting paper below!

Why do we tolerate loud motorcycles? I can't walk around with a crazy loud speaker, yet we allow people to have (and companies to sell) insanely loud machines that torture everyone else. They can be made quiet or silent...we should start heavily taxing and fining noise polluting machines.

Very cool work! Fun to see an algorithm inspired by MAP-Elites help advance numerous open problems in math, computer science, AND improve Google's production infrastructure (data centers, TPU design, and AI training)! (Paraphrasing their paper and tweet).

I greatly enjoyed “The Spectrum of AI Risks” panel at the Singapore Conference on AI. Thanks @teganmaharaj.bsky.social for great moderating, Max Tegmark for the invitation, and the organizers and other panelists for a great event! PS. Do I really have sad resting panel face? 😐

I'm giving a talk at 11:30 today in the #ICLR World Models workshop. "Open-ended Agent Learning in the Era of Foundation Models and Foundation World Models." Drop by if you are interested! sites.google.com/view/worldmo...

Tom @rockt.ai did a great job in his #ICLR2025 keynote on open-endedness of explaining the ideas we are all so passionate about. A huge thanks for the kind words and for featuring our work, including Jenny Zhang's OMNI. cc @kennethstanley.bsky.social @joelbot3000.bsky.social

Very excited for this keynote by @_rockt! Awesome to see open-endedness go from a niche (😉) area to a keynote at #ICLR ! 🌱🌿🌳🌲🍀🌍✨ 📈 🧬🧪 cc @joelbot3000.bsky.social @kennethstanley.bsky.social

Since the dawn of my career I've heard scientists joke after every AI advance “Now we just need to get AI to write the paper. Now The AI Scientist does all the research AND writes the paper! Wild times! 🧪🔬👩‍🔬🤖

Introducing The AI Scientist-v2, which produced the 1st fully AI-generated paper to pass peer review at a workshop level (at #ICLR2025) ‼️ Tech Report: pub.sakana.ai/ai-scientist... GitHub: github.com/SakanaAI/AI-... This work is a proud collaboration between Sakana AI, UBC, and Oxford University.

Collecting a diamond in #Minecraft is “a very hard task”, says Dr. Jeff Clune (@jeffclune.com) of @cs.ubc.ca, who was part of a team that trained a program to find diamonds using videos of human play2. “This represents a major step forward for the field.” www.nature.com/articles/d41...

Trump is the greatest gift to China in history.

Awesome work. Great to see Dreamer in Nature! Congrats Danijar Hafner et al. Nature news article on it with quotes from yours truly. Below is the quote I provided that did not make it into the article. www.nature.com/articles/d41...

I may be tired and a little hoarse, but as I said again and again on the Senate floor, this is a moment where we cannot afford to be silent, when we must speak up.

When whites and minorities drive at identical speeds (according to objectively measured data from Lyft) Florida police are 24-33% more likely to issue speeding citations to minority drivers and charge them 23-34% greater fines. These are not small effects!

It is an honor to receive a Killam Accelerator Research Fellowship. Thank you to everyone involved!

What a remarkable milestone in history! Since at least 2010 I heard the joke amongst AI scientists "Now if only we could get AI to write the paper." Fast forward 14 years and it's possible! Exciting times! 🧪 🧫 🔬 🚀 📈

Your chance to work with one of the all time great minds in science! 🚀🚀🚀🚀🚀🚀

Honored to serve on the Government of Canada's Safe & Secure AI Advisory Group to keep it "well informed on risks associated with AI systems" w/ AI luminaries, e.g. @yoshuabengio.bsky.social Joelle Pineau, Elissa Strome, @davidduvenaud.bsky.social & many others ised-isde.canada.ca/site/advisor...

My guest lecturing for the Stanford CS course Self-Improving AI Agents. The talk is online, titled "Open-ended Agent Learning in the Era of Foundation Models" (w/ more emphasis on the AI Scientist and ADAS than prior versions). Thanks profs @Azaliamirh & @achowdhery! www.youtube.com/watch?v=EZBu...

It was an honor to be on Quirks and Quarks (the CBC science show) with @cong-ml.bsky.social talking about The AI Scientist and the impact of AI on science. Science is being transformed by the AI revolution cbc.ca/listen/live-...

Introducing Automated Capability Discovery! ACD automatically identifies surprising new capabilities and failure modes in foundation models, via "self-exploration" (models exploring their own abilities). Led by @cong-ml.bsky.social & @shengranhu.bsky.social 🔬🤖🧠🔎 [1/9]

🤔 What if AI could explain its thought process before taking action? Our latest ANDERS blog covers Vector Researchers @jeffclune.com & @shengranhu.bsky.social’s wrk on "thought cloning" - teaching AI to express its reasoning process in language we can understand vectorinstitute.ai/thought-clon...

Nice to see international collaborations of this scale on AI safety! Also nice that our work (four papers) contributed to the discussion (full disclosure: that's both on safety, but also on creating powerful general agents, including mentions of The AI Scientist and SIMA). tldr; it's complicated

Operator is the descendant of VPT (work our team did at OpenAI on learning to use a mouse+keyboard to perform long-horizon tasks (demonstrated in Minecraft). As we wrote in that blog post, VPT is "a step towards general computer-using agents." Released ~2.5 years later. openai.com/index/introd...

Research Idea: Predict abstract properties of text in addition to next-word prediction as helpful auxiliary losses. Has this been tried? (1/n)

Our UBC lab + collaborators had 3/3 papers accepted to #ICLR2025! 🚀🚀🚀🍻🍻🍻💥💥💥👊👊👊💡💡💡 Thanks @iclr-conf.bsky.social ACs, reviewers, and organizers. 🙏 See everyone in Singapore! 🇸🇬

A copy of Tim's @rockt.ai has arrived. As I say on the back cover, I read it cover to cover and recommend it to those looking to ramp up on modern AI. Congrats and thanks Tim!

Many comments from @geoffreyhinton & myself in this nice CBC article. Do you disagree with us? "The stakes are extremely high. If we get this right, we get tremendous upside. And if we get this wrong, we might be talking about the end of human civilization." www.cbc.ca/news/science...

It's an honor that The AI Scientist is #1 on this list! www.linkedin.com/feed/update/... Congrats @chris-lu.bsky.social @cong-ml.bsky.social @RobertTLange @hardmaru.bsky.social @jfoerst.bsky.social

I strongly believe the development of superintelligence is inevitable. I do not believe humanity has the ability to not invent it, given its economic, scientific, and military value. Many people say things like “we should not build AGI” without realizing that outcome is virtually impossible. 1/

I greatly enjoyed this interview on MLStreetTalk It covers a wide range of topics I am interested in (open-endedness, AI-GAs, AI safety, etc.), includes my scientific origin story, and has great visuals edited in. Thanks Tim Scarfe! www.youtube.com/watch?v=mw5W...

I am one of the PIs you could apply to work with via this excellent program. Please consider applying!

I guess AI and human scientists stand on the shoulders of the same giants.

NO WAY! This is nearly EXACTLY one of the paper ideas The AI Scientist came up with! It was my favorite idea it generated, & made us very impressed with its creativity & good taste in an ML paper proposal. Cool to see humans agree! The humans def executed better though. x.com/BrantonDeMos...

Dream finish to #NeurIPS2024: a dinner with Ted Chiang (one of my favorite writers), the excellent @alisongopnik.bsky.social, many amazing open-ended researchers (including @cedcolas) & other excellent people. Thanks to the tremendous IMOL organizers! Until next time! 🦎🧬🦖🦣

Lots of interest in ADAS! Thanks everyone, and congrats Shengran Hu and @cong-ml.bsky.social! 🚀🚀🚀

Jenny did a great job presenting OMNI-EPIC during the IMOL Oral. Congrats Jenny Zhang! arxiv.org/abs/2405.15568

A huge congratulations to Shengran Hu and @cong-ml.bsky.social on ADAS winning an Outstanding Paper Award at the #NeurIPS2024 OWA workshop!! And nice jumps everyone!! 🦘 🦘 🦘

This is the newest work out of our lab. Please stop by if you're curious to learn about our latest, which I'm very excited about!